apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
14,136 Commits 1 Branch 0 Tags
5ae0bd99bcaff3e69b80333dcfb57388e3364f55
Commit Graph
14136 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Ryan Curtin 5ae0bd99bc Loosen tolerance further. 
This is better-tested than my last attempt and appears to have a very low failure rate.
 2016-02-01 10:23:06 -08:00
Ryan Curtin e7196b502a Update version number. I forgot to do this! 2016-02-01 08:43:10 -08:00
Ryan Curtin 725f7d01aa Fix bug where arma_config.hpp wouldn't be installed. 
This would happen when the user configured once and immediately installed.  The
issue was that MLPACK_SRCS had arma_config.hpp appended to it too late, and
didn't exist at the time of the first scan of files.  When CMake was run a
second time, arma_config.hpp existed and thus would get picked up by the scan
and added to MLPACK_SRCS.  This is a bug I didn't run into because I don't build
from a new environment every time...
 2016-02-01 08:37:15 -08:00
Ryan Curtin 47d8a6aa4e Wrap to 80 columns. 2016-02-01 08:37:15 -08:00
Ryan Curtin 3b3a8a1072 Merge pull request #512 from theranger/master 
Use CMAKE_CURRENT_SOURCE_DIR during files generation.
 2016-02-01 07:27:13 -08:00
Ryan Curtin eb2114e0ac Widen tolerances; PositiveDefiniteConstraint seems to change things. 2016-02-01 09:28:18 -05:00
marcus 8b4bac7354 Remove unused parameter. 2016-02-01 00:58:45 +01:00
marcus 2f105ecea1 Use the modified sparse autoencoder test file. 2016-02-01 00:07:00 +01:00
marcus 443ecdcd35 Use the sparse autoencoder class for the sparse autoencoder test. 2016-01-31 23:54:32 +01:00
marcus f34ae33e2c Remove unnecessary element function for the convolutional network and add sparse autoencoder training routine that makes use of the batch size. 2016-01-31 23:35:00 +01:00
marcus 1896937d03 Minor style and formatting changes. 2016-01-31 23:29:51 +01:00
marcus 4c5c7739d4 Add the sparse autoencoder to the network traits. 2016-01-31 23:22:54 +01:00
marcus 4ad39f8651 Move the sparse autoencoder into a separate folder and remove legacy code. 2016-01-31 23:21:11 +01:00
The Ranger de42a6aaa6 Use CMAKE_CURRENT_SOURCE_DIR during files generation. 2016-01-30 18:43:10 +02:00
Marcus Edel 33082f0bf0 Merge pull request #451 from stereomatchingkiss/refactor_sparse_autoencoder 
Improve speed of SparseAutoencoder and make it more flexible.
 2016-01-30 13:34:57 +01:00
Ryan Curtin 54b2906d83 A better positive definite constraint strategy. 
Instead of just checking if the Cholesky factorization succeeds (which it might
and it may still be inaccurate), use the condition number of the matrix to
prevent numerical instability.  I suspect this may be somewhat slower than
before, though, but this is the price of stability...
 2016-01-29 15:32:47 -05:00
Ryan Curtin 1f6db3049b Check number of columns. 2016-01-29 09:55:38 -08:00
Ryan Curtin ab37969687 Better error output for ARFF. 
It wasn't properly printing the error for missing values.
 2016-01-29 08:22:22 -08:00
Ryan Curtin a7a05d8fe4 Fix typo: we should be checking HAVE_LAPACK not LAPACK_FOUND. 
LAPACK_FOUND is only set if MKL/ACML/etc. isn't found, but HAVE_LAPACK is set in
both situations.
 2016-01-27 07:34:58 -08:00
Ryan Curtin 3561ad7b90 Refactor to allow arbitrary types for Range. 
But for reverse compatibility, make RangeType<T> be math::Range, and upon 3.0.0
we'll refactor so RangeType is Range.  (That will break reverse compatibility.)
 2016-01-25 11:52:03 -05:00
stereomatchingkiss 1a233b4d7a remove useless preprocessor 2016-01-25 22:47:48 +08:00
stereomatchingkiss e481a0a29c change test suite name 2016-01-25 22:19:13 +08:00
stereomatchingkiss 4a9ddb2eb9 remove useless header file 2016-01-25 22:15:18 +08:00
stereomatchingkiss c34e9d3c8b remove useless variable 2016-01-25 22:10:22 +08:00
stereomatchingkiss 9721cb7a41 fix bug--lvalue cannot bind to rvalue 2016-01-25 21:49:31 +08:00
stereomatchingkiss 12ea51f937 do not make outputLayer as reference but allow user to access the outputLayer from public api of FFN 2016-01-25 16:46:30 +08:00
stereomatchingkiss 94a3de9eee use rvalue reference to pass the value of outputLayer 2016-01-25 16:41:31 +08:00
stereomatchingkiss 1c2af05678 change typenmae from FFN to FFNet 2016-01-25 15:55:58 +08:00
stereomatchingkiss c1d23bc316 change template name due to confliction 2016-01-25 14:30:16 +08:00
stereomatchingkiss 61d013a085 change function name and fix type error 2016-01-25 13:45:27 +08:00
stereomatchingkiss 18aad1ca08 use less generic solution to find out the value type of InputDatatType because travis build cannot pass 2016-01-25 13:21:46 +08:00
stereomatchingkiss d697ad8dec use less generic solution to find out the value type of InputDatatType because travis build cannot pass 2016-01-25 13:14:56 +08:00
stereomatchingkiss 6e2710f9e9 remove non exist and useless headers 2016-01-25 12:05:02 +08:00
stereomatchingkiss e98bc7729a first commit 2016-01-25 04:36:15 +08:00
stereomatchingkiss 5e353ff3f4 fix bug--forgot to move some values 2016-01-25 03:16:03 +08:00
marcus a279679d42 Remove self assignment. 2016-01-23 13:08:26 +01:00
Ryan Curtin 0c9829c9ed Fix issue that caused hanging tests: use the intended tolerance. 2016-01-22 13:27:54 -05:00
Ryan Curtin de6c7b6e04 Print status while loading. 2016-01-22 09:50:05 -05:00
Ryan Curtin e6bbde6949 Merge pull request #505 from andyfangdz/patch-1 
Update GetParam<>(parameterName) per marcos
 2016-01-22 05:59:22 -08:00
Andy Fang cd15763292 Update GetParam<>(parameterName) per marcos 
```
PARAM_STRING("input_model_file", "File containing existing model (parameters).",
    "m", "");
PARAM_STRING("output_model_file", "File to save trained logistic regression "
    "model to.", "M", "");
```
v.s.
```
const string inputModelFile = CLI::GetParam<string>("input_model");
const string outputModelFile = CLI::GetParam<string>("output_model");
```
 2016-01-21 22:09:44 -05:00
Ryan Curtin 941fd2008a Widen tolerances so tests succeed more. 2016-01-19 14:10:55 -05:00
Ryan Curtin 4bbc39aa39 Print correct parameter name. 2016-01-19 13:31:32 -05:00
Ryan Curtin b29997d08d Merge pull request #504 from andyfangdz/patch-1 
Fix confusion in error message
 2016-01-19 10:30:54 -08:00
Andy Fang d72958ecda Fix confusion in error message 
The error message was "missing output_model",
while the variable used for detecting it was "outputFile",
which refers to the testing output, rather than the model output.
 2016-01-19 13:22:15 -05:00
Ryan Curtin 6f8d7a19e3 Merge pull request #502 from barak/master 
COPYRIGHT.txt, repeated field and lintian tag copyright-refers-to-dep…
 2016-01-19 05:41:03 -08:00
Barak A. Pearlmutter ee57276625 COPYRIGHT.txt, repeated field and lintian tag copyright-refers-to-deprecated-bsd-license-file 2016-01-19 08:44:58 +00:00
Ryan Curtin e6d730836c Merge pull request #501 from barak/master 
spelling
 2016-01-18 15:04:24 -08:00
Barak A. Pearlmutter b39417f511 spelling 
Gbp-Pq: Name 0002-spelling.patch
 2016-01-18 16:57:54 +00:00
Ryan Curtin c5050982ef Force a minimum value on the diagonal. 
This is necessary so that the inverse covariance doesn't explode.  If the
covariance is sufficiently small, then the PDF of the Gaussian will condense
towards a delta function, and you will end up with inf's in your probabilities,
which turn into NaNs when normalized, and then your entire model turns into
NaNs.

Note that the PositiveDefinite constraint can be replaced with the NoConstraint
class for speed if necessary.
 2016-01-13 14:45:42 -05:00
Ryan Curtin 6c42661c33 Handle an edge case when all the points are the same. 2016-01-13 11:12:05 -05:00